Textual data majorly reflects objective and subjective human specific knowledge. Focusing on big data in industrial and operation management, the value of textual data is oftentimes undermined. Optimal use of data reinforces the integrative modeling and analysis of RAM (Reliability, Availability, Maintainability). Data-driven reliability engineering and maintenance management, gains benefit from textual data, especially for identifying unknown failure modes and causes, and solving problems. The scientific challenge is how to effectively discover knowledge from text data and convert it into automated processes for inferential reasoning, predicting and prescribing. This paper outlines how the reliability-centered maintenance in production systems can be improved by explicating and discovering human-specific knowledge from maintenance reports and related textual documents. Hence, a theoretical model for text understanding is proposed, which is demonstrated as a proof-of-concept demonstrator using real world manufacturing datasets. The text understanding model is represented by a three-dimensional matrix comprising three indexes, i.e. text readability, word associations within texts as well as sentiment. The implementation of the model as a software prototype involves using text mining techniques and machine learning algorithms. This paper emphasizes on the importance of knowledge extraction from text in the context of industrial maintenance, by demonstrating how an increased value of text understandability of maintenance reports correlates to an early stage detection of failure, the reduction of human failures and leads to an immense improvement of explication of human knowledge.